embodied AI

Embodied AI integrates machine learning with physical hardware to create agents that perceive, move, and interact within the real world.

This field moves intelligence beyond static datasets into dynamic environments. By combining computer vision, reinforcement learning, and robotics, systems like Google’s RT-2 or Boston Dynamics’ Atlas translate high-level reasoning into precise physical motor commands. The goal is a closed-loop system where the agent learns through sensorimotor interaction (trial and error) rather than just passive observation. This shift enables hardware to navigate complex spaces, manipulate objects with human-like dexterity, and perform unstructured tasks in homes and factories.

https://ai.stanford.edu/blog/embodied-ai/